DEER: Disentangled Mixture of Experts with Instance-Adaptive Routing for Generalizable Machine-Generated Text Detection

2025-11-02

AI Summary

Weak generalization under cross-domain settings and training-inference inconsistency due to missing domain labels hinder machine-generated text (MGT) detection. To address this, we propose a decoupled Mixture-of-Experts (MoE) architecture that explicitly separates domain-specific experts from a universal shared expert. We further design a reinforcement learning–based instance-adaptive routing mechanism to enable dynamic, unsupervised expert selection without domain supervision. This design jointly preserves domain discriminability and feature universality, thereby mitigating distribution shift. Evaluated on five in-domain and five cross-domain benchmarks, our method consistently outperforms state-of-the-art approaches: it improves F1 score by 1.39% (in-domain) and 5.32% (cross-domain), and accuracy by 1.35% and 3.61%, respectively. The results demonstrate substantial gains in cross-domain robustness and practical applicability of MGT detectors.

Technology Category

• Machine Learning: Mixture of Experts (MoE)
• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Application Domains: Mobility, Driving & Flight

Application Category

• Search and Retrieval-Augmented AI: Vertical and domain-specific search
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Detecting machine-generated text (MGT) has emerged as a critical challenge, driven by the rapid advancement of large language models (LLMs) capable of producing highly realistic, human-like content. However, the performance of current approaches often degrades significantly under domain shift. To address this challenge, we propose a novel framework designed to capture both domain-specific and domain-general MGT patterns through a two-stage Disentangled mixturE-of-ExpeRts (DEER) architecture. First, we introduce a disentangled mixture-of-experts module, in which domain-specific experts learn fine-grained, domain-local distinctions between human and machine-generated text, while shared experts extract transferable, cross-domain features. Second, to mitigate the practical limitation of unavailable domain labels during inference, we design a reinforcement learning-based routing mechanism that dynamically selects the appropriate experts for each input instance, effectively bridging the train-inference gap caused by domain uncertainty. Extensive experiments on five in-domain and five out-of-domain benchmark datasets demonstrate that DEER consistently outperforms state-of-the-art methods, achieving average F1-score improvements of 1.39% and 5.32% on in-domain and out-of-domain datasets respectively, along with accuracy gains of 1.35% and 3.61% respectively. Ablation studies confirm the critical contributions of both disentangled expert specialization and adaptive routing to model performance.

Problem

• Detecting machine-generated text across domain shifts
• Learning domain-specific and domain-general patterns simultaneously
• Adapting expert selection dynamically without domain labels

Innovation

• Disentangled mixture-of-experts module captures domain-specific and cross-domain features
• Reinforcement learning-based routing mechanism dynamically selects experts per instance
• Two-stage architecture bridges train-inference gap caused by domain uncertainty